This application has three specific aims dealing with the design and analysis of studies into population pharmacokinetics (PK) and pharmacodynamics (PD), with application to cancer chemotherapy. The first specific aim proposes expanding upon an earlier Bayesian analysis of hematology data collected as part of a Cancer and Leukemia Group B (CALGB) Phase I study. In particular, it is proposed to develop a Bayesian semiparametric model, one flexible enough to allow heterogeneity and overdispersion in the distribution of patient-specific parameters (random-effects distribution), as well as non-parametric regression on patient-specific covariates. Extensions will also include models for multiple longitudinal outcomes. The investigators will reanalyze the CALGB Phase I study and evaluate differences from the applicant's earlier analysis. The second specific aim is to develop a predictive model relating hematologic toxicity to patient characteristics. The investigators will develop a Bayesian hierarchical metamodel and apply it to data from two completed CALGB studies: the Phase I study mentioned in aim one and a large Phase I trial of adjuvant chemotherapy for women with stage II breast cancer. The third aim proposes new methodology for solving optimal design problems built around PK/PD models, honestly accounting for uncertainty in the estimation and prediction in the PK/PD models. In particular, the investigators will apply decision-theoretic considerations to develop a rational strategy for picking times to sample patient plasma in population pharmacokinetic and pharmacodynamic studies. They will evaluate potential savings, compared to current limited sampling strategies, via simulation studies under various presumed pharmacokinetic and pharmacodynamic models reported in the literature.